What M5 Ultra changes
Apple announced a new Mac Studio with M5 Ultra on August 25. The top configuration has a 36-core CPU, an 80-core GPU, 512GB of unified memory, and 1.2TB/s of memory bandwidth.
M3 Ultra already supported 512GB. M5 Ultra keeps that capacity and raises memory bandwidth by 50% to 1.2TB/s. Apple also claims up to 4.3 times the peak AI compute performance.
The CPU and GPU use the same memory pool, avoiding the separate GPU-memory ceiling that often limits local model deployments.
Does Leiolai run on an M5 Ultra?
Yes. Leiolai is available on Mac, including Mac Studio with M5 Ultra.
Leiolai is an intelligence that runs on consumer hardware and pays users for the compute they provide. Leiolai 1 is available in the Leiolai app and through the Leiolai API.
Run 975B Inkling on one M5 Ultra
Thinking Machines publishes the 975B model as thinkingmachines/Inkling. It has 975 billion total parameters, 41 billion active parameters per token, and a one-million-token context window. It supports text, images, and audio.
The Mac-compatible unsloth/inkling-GGUF:UD-Q3_K_XL build is 433GB in weight files.
Can you run full precision Inkling on four M5 Ultras?
The original thinkingmachines/Inkling checkpoint totals 1.9TB. Four 512GB M5 Ultra Mac Studios can be linked over Thunderbolt 5 to create a 2TB memory pool.
Apple says four linked Mac Studios can deliver up to three times the inference performance of one.
The M5 Ultra Mac Studio starts at $5,499 in the United States and ships September 22. Apple says the 512GB configuration will follow in late October.
Sources Apple’s Mac Studio announcement, Apple’s M5 Ultra technical overview, Thinking Machines’ Inkling overview, the official Inkling model card, and Unsloth’s Inkling GGUF repository.